Segmenting everything, everywhere, all at once

Segmentation is a popular task in the field of computer vision, which requires to label every pixel of an image and video. Recently, foundation models emerged that promise to tackle one of the biggest limitations of previous work - the ability to segment arbitrary images and videos. In this blockseminar, we will explore current trends and techniques in segmentation, which will pave the road for the visual perception of machines in the future.

Checkout https://segment-anything.com for visual examples.

General information

Date: Blockseminar 20.03.2024 - 21.03.2024

Lecturers: Prof. Dr. Laura Leal-Taixé and Mark Weber.

Prerequisites

• Introduction to Deep Learning (I2DL) (IN2346)
• Strong mathematical foundations
• Knowledge of Transformer and CNNs

Preferred: Computer Vision 3: Detection, Segmentation and Tracking The course can also be taken in parallel next semester.

Course matching

Students are supposed to both

• register to the Matching-System. See the Matching-System FAQ for more details.
• write an email to Mark, subject “Segmentation Seminar Application - your name” and include your Deep Learning background (I2DL etc.) and motivation to sign up for this seminar.

After the final matching is announced, we will send an email to all participants with further information. We will have space for up to 12 students.

Papers

We will propose a list of important papers relevant for the seminar. In addition, students are encourage to propose papers in this domain that that they are interested in and we will match those early in the winter semester.

Schedule

• 20.03.2024 Full-Day Seminar presentation

• 21.03.2024 Full-Day Seminar presentation

• 19.04.2024 Tentative deadline for written reports